TL;DR
This paper presents a novel DEM fusion scheme that leverages information entropy to determine the optimal resolution and weighted averaging coefficients for combining SRTM1, ASTER GDEM V3, and AW3D30. The approach significantly improves DEM accuracy, particularly in complex mountainous terrain, by adaptively weighting data sources based on slope.
What Problem Does This Paper Address?
Digital Elevation Models (DEMs) are crucial for a wide range of geospatial applications, from hydrology to geological surveying. However, existing global DEMs like SRTM, ASTER GDEM, and AW3D30, while valuable, have inherent limitations. SRTM can suffer from voids due to terrain and vegetation, while ASTER GDEM and AW3D30, derived from photogrammetry, can be affected by cloud cover. Simply merging these datasets doesn't guarantee optimal results. This paper addresses the challenge of creating a more accurate and reliable DEM by developing a sophisticated fusion strategy that accounts for the strengths and weaknesses of each source, especially considering the varying topographic characteristics of a study area.
Methods
The authors propose a two-pronged approach:
- Optimal Resolution Determination: They utilize the concept of information entropy to quantify the terrain information contained within a DEM. By calculating the information entropy of key topographic factors (slope, aspect, plane curvature, profile curvature) across various DEM resolutions, they identify the optimal resolution that maximizes terrain representation without introducing excessive data redundancy. This is achieved by resampling the input DEMs to different resolutions and analyzing how entropy changes.
- Weighted Average Fusion with Slope-Based Weighting: To achieve the best fusion, they employ a weighted average fusion technique. Crucially, instead of applying a single set of weights across the entire study area, they divide the area into nine slope classes. For each slope class, they use a traversal search method to find the optimal weighting coefficients for SRTM1, ASTER GDEM V3, and AW3D30 that minimize the Root Mean Square Error (RMSE) when compared to a high-resolution reference DEM. This adaptive weighting acknowledges that different DEMs perform better in different topographic conditions.
Datasets
The study utilizes the following global DEM datasets:
- SRTM1 DEM: Shuttle Radar Topography Mission, resolution of 1 arc-second.
- ASTER GDEM V3: Advanced Spaceborne Thermal Emission and Reflection Radiometer Global Digital Elevation Model, resolution of 1 arc-second.
- AW3D30 DEM: ALOS World 3D, resolution of 30 m.
A reference DEM with a 5 m resolution, acquired via LiDAR, was used for validation and accuracy assessment.
Key Insights and Ideas
- Information Entropy as a Resolution Metric: Using information entropy provides a quantitative and objective way to determine the optimal DEM resolution that balances terrain detail with data efficiency.
- Adaptive Fusion is Key: The performance of DEMs varies significantly with topography. A one-size-fits-all fusion approach is suboptimal; weighting DEMs based on slope dramatically improves accuracy.
- AW3D30's Strength in Complex Terrain: The study highlights AW3D30's superior accuracy in representing complex mountainous terrain, leading to its higher weighting in steeper areas.
- SRTM1's Utility in Flat Areas: Conversely, SRTM1 shows higher relative accuracy in very gentle slopes (0°-5°), contributing more significantly to the fused DEM in these regions.
Key Results
- Optimal Resolution: The optimal DEM resolution for the study area (Taiyuan City) was determined to be 40 m.
- Accuracy Improvement: The fused DEM demonstrated significant error reductions compared to the individual input DEMs:
- Mean Absolute Error (MAE) decreased by 33.8% (SRTM1), 57.9% (ASTER GDEM V3), and 11.5% (AW3D30).
- Standard Deviation (STD) decreased by 36.3% (SRTM1), 54.6% (ASTER GDEM V3), and 1.4% (AW3D30).
- Root Mean Square Error (RMSE) decreased by 32.8% (SRTM1), 54.2% (ASTER GDEM V3), and 9.7% (AW3D30).
- Topographic Parameter Accuracy: The fused DEM also led to improved accuracy in derived topographic parameters like slope, profile curvature, and TPI.
- Complex Terrain Performance: The fused DEM showed particularly high accuracy in complex mountainous validation areas, outperforming individual DEMs.
Limitations and Future Work
The authors acknowledge that the current fusion scheme primarily considers the correlation between slope and elevation. Future work could explore multi-dimensional correlation analysis by incorporating other terrain-related factors such as vegetation cover and aspect to achieve even more accurate fusion results. They also suggest investigating regression relationships between topography, landforms, vegetation cover, and DEM accuracy.
Summary updated